How to write a clinical case for teaching

The teacher has fifteen minutes before class and writes the case from memory: "Man, 58 years old, crushing chest pain radiating to the left arm, sweating." In the room, the class answers the diagnosis in three seconds and the discussion ends there. The case was correct, but it taught nothing: it gave away the answer in the first line.

NEXO is a daily medical diagnosis game: one clinical case a day, six progressive clues, and ICD-10 proximity feedback across five levels.

This guide is for those who write cases: teachers, preceptors, monitors, academic leagues and students who put together material for the class. It shows how to start from the objective, how to dose the clues, how to calibrate the difficulty, how to review and how to use artificial intelligence without losing control of the text.

Key points

• A teaching case starts from what the reader needs to decide, not from the disease.
• Well-constructed vignettes measure clinical practice closer to the observed consultation than chart review, according to validation studies.
• The key features approach recommends concentrating the case on the critical decisions, where errors usually happen.
• The patient's story needs immediacy and humanity, and the author must watch for biases and stereotypes in the text.
• AI-generated vignettes were considered useful by physician reviewers, but they asked for correction of details and specialist supervision.

What a clinical teaching case is

A story with data and a question

In an article about case studies in research and education, Trisha Greenhalgh describes educational cases as real-world topics presented as stories illustrated by data, which lead students to discuss the subject from different angles (Greenhalgh, 2025). The author also warns that many reports called case studies are too short and superficial (Greenhalgh, 2025).

In medicine, a teaching case is a clinical story written to prompt a decision. It can be a short vignette in an exam, a multi-session problem, a case in clues or the script of a simulation. In all of them, the author chooses what to show, in what order and what to leave out.

Teaching case and case report

The case report published in a journal describes a patient who actually existed, usually because it is rare or because it teaches something new. The teaching case is constructed: it can combine data from different patients, change details and be fictional, as long as the clinical presentation is plausible. Its commitment is to the learning objective, not to fidelity to a person.

Where this comes close to other topics on the blog

Solving cases, alone or in a group, is in the guide on case-based learning. Acting out the case with a colleague in the patient role is in the guide on clinical simulation. Here the perspective is that of the person who writes.

Start from the objective, not from the disease

What decision should the reader make?

The most common mistake is to start from the disease and write everything known about it. The case becomes complete and obvious. The reverse path works better: first decide what the reader should learn to do, for example "distinguish two similar causes by the history" or "notice that a finding changes the priority", and only then choose the condition that forces that decision.

Focus on the critical decisions

The key features approach was proposed in 1984 as a way to assess clinical decision-making by concentrating each case on the critical steps, the ones where errors happen most (Bordage and Page, 2018). A review of more than three decades of studies gathered validity evidence for this approach at all levels of training and in various exam formats (Bordage and Page, 2018).

Even for those who write class cases rather than exam cases, the idea works: each case should have one or two decisions that matter, and the whole text works so that the reader reaches them. The guide on clinical reasoning games shows how key features appear among other formats.

Write the objective in one sentence

Before the first line of the case, write: "By the end, the reader should be able to...". If the sentence does not fit, the case is trying to teach too many things. Split it in two.

How to dose the clues

From the vaguest to the most specific

In a case designed to train reasoning, the order of the information is the content itself. Start with data that supports several hypotheses, such as age, complaint and context, and move on to data that separates the hypotheses. The clue that solves the case comes last, or near the end. The guide to differential diagnosis shows how the reader uses this progression.

Every clue has to change something

Reread the case and ask, line by line: if I remove this information, does the reader's list of hypotheses change? If it does not change, the line is decoration or distraction. Distractions have their place, because patients also bring irrelevant data, but they should be chosen, not left in the text by accident.

Keep the name out of the description

Words like "typical", "classic" or terms that exist in only one disease give away the answer. Describe what the patient reports and what is observed, in the language of someone examining the patient, and leave the interpretation to the reader. The guide to signs, symptoms and history helps separate data from interpretation.

Give the patient humanity

A guide on writing cases for problem-based learning recommends giving the story immediacy and making the reader see the patient as a person (Bergemann et al., 2025). One line about the patient's work, family or worry makes the case more memorable and reminds the reader that the decision affects someone.

Difficulty, ambiguity and answer

Vignettes measure what is done

Clinical vignettes are not just a classroom exercise. In a validation study in primary care clinics, the same physicians were assessed in three ways: by standardized patients who arrived unannounced, by review of the records of those consultations and by vignettes that reproduced exactly the same presentations (Peabody et al., 2000). The vignette scores were consistently closer to the standardized patient scores than the record review scores (Peabody et al., 2000).

A larger study, with 116 physicians, confirmed the pattern: the quality measured by standardized patients was 73%, by vignettes 68% and by record review 63% (Peabody et al., 2004). For writers, the message is encouraging: a well-made vignette comes close to what the physician would do with the patient in front of them.

Ambiguity on purpose

Some cases call for more than one path. To train decisions in uncertain situations, the script concordance test uses short cases in which a new piece of data is presented and the reader says whether the hypothesis becomes more or less likely (Fournier et al., 2008). The guidelines for building this format describe how to write the case, the hypothesis and the new data, and how to compare the answers with those of a panel of experts (Fournier et al., 2008).

The lesson for the author: if the goal is to teach how to deal with doubt, ambiguity must be planned, and the reference answer must come from more than one experienced person.

Choose the answer format

An open answer forces the reader to generate the hypothesis; multiple choice options ask the reader to recognize it. Both have their uses, but they train different things. If the goal is diagnostic reasoning, prefer to have the reader write the hypothesis before seeing any option.

Write the explanation along with the case

The case ends with the explanation. Write why the answer is the expected one, why the plausible alternatives fall short and which clue separated one from the other. The guide to feedback in clinical cases shows what makes an explanation useful for someone who got it wrong.

Review: the case before it goes to the class

Ask someone who knows the subject to review it

The guide by Bergemann and colleagues sets aside a step for incorporating expert opinion into the content (Bergemann et al., 2025). In practice, ask a colleague to solve the case without knowing the answer. If they arrive by another path, or do not arrive at all, the text needs adjusting.

Look for bias in the text

The same guide advises the author to avoid unintended bias and microaggressions (Bergemann et al., 2025). Ask: why does this patient have this age, this sex, this profession, this origin? If the characteristic is there only to suggest the diagnosis through a stereotype, the case teaches the stereotype along with it.

Test it with a student

A student at the class level shows whether the difficulty is right. Ask them to think aloud while reading and note where they hesitate. If the answer comes on the first line, the case is easy; if nobody gets there, a clue was missing.

Artificial intelligence in case writing

What the reviewers thought

In a Japanese study, physicians with experience in education evaluated 18 clinical vignettes generated by a language model; 71 of the 73 invited responded, or 97% (Takahashi et al., 2024). Educational usefulness received a mean of 3.55 on a five-point scale, and mean diagnostic difficulty was 2.34 (Takahashi et al., 2024). The reviewers asked for more coherence between the clinical information, more realism in the details and patient-centered presentations, and the authors concluded that use depends on specialist supervision (Takahashi et al., 2024).

Almost impossible to tell who wrote it

In a residency program, 33 residents and preceptors answered vignette-based questions, half written by people and half by a language model with structured instructions (Jackson et al., 2025). They correctly identified the authorship a median of 39.1% of the time, and the question accuracy rate was similar in both groups, 62.3% and 64.4% (Jackson et al., 2025).

How to use it without losing control

AI speeds up the draft, but the objective, the order of the clues and the review stay with the author. Give the model the learning objective and the critical decision, ask for several versions, choose one and review every piece of data as you would a colleague's text. The guide on artificial intelligence in clinical reasoning discusses the limits of these models.

A case in clues as a model: NEXO

The NEXO case of the day follows the logic described in this guide. It arrives in up to six progressive clues, from the vaguest to the most specific, and the reader writes their own hypothesis, looking for the diagnosis in ICD-10. With each guess, proximity across five levels shows whether the hypothesis landed in the right chapter, group or category.

The cases are fictional with a real clinical presentation, and each one goes through an audit against the NEXO Core knowledge base before publication. For those who write cases, playing a few and noting where each clue changed your hypothesis is a good reverse-engineering exercise. The Archive gathers the previous cases, and how the modes are unlocked is on the NEXO economy page. The guide what NEXO is explains the whole game.

Physicians who want to review cases in their specialty take part through NEXO Curator. Teachers and coordinators who want to use the cases with their class can learn about NEXO Closer, and researchers can propose studies through NEXO Researcher. Those who create health content on social media take part through NEXO Creators.

Frequently asked questions

Where should you start when writing a clinical case for teaching?

With the objective: write in one sentence what the reader should be able to decide at the end, and only then choose the clinical picture.

How many clues should a case have?

Enough that each one changes the list of hypotheses. If a piece of information can be removed without changing anything, it is decoration or a deliberate distraction.

Do clinical vignettes measure real practice?

In validation studies, vignette scores came closer to those of standardized patients than to those of chart review (Peabody et al., 2004).

Can I use artificial intelligence to write cases?

You can, as a draft. Physician reviewers found the vignettes useful, but asked for corrections and specialist supervision (Takahashi et al., 2024).

How do you keep the case from giving away the answer?

Remove words like "typical" and "classic", describe what is observed instead of the name of the finding and leave the decisive clue for the end.

What does the NEXO case offer to those who write cases?

It shows the progression of up to six clues with an open answer and feedback through proximity in ICD-10, a format you can study and adapt.

Checklist for writing a case

1. Write the objective in one sentence.
2. Choose one or two critical decisions.
3. Order the clues from the vaguest to the most specific.
4. Cut what does not change the list of hypotheses.
5. Remove the words that give away the diagnosis.
6. Give humanity to the patient and watch for stereotypes.
7. Write the explanation along with the case.
8. Ask a colleague for a review and test it with a student.
9. Review everything the AI writes.

Want to see a case in clues from the inside? Play the case of the day on NEXO.

References

1. Greenhalgh T. Case studies: a guide for researchers, educators, and implementers. BMJ Med. 2025. https://pubmed.ncbi.nlm.nih.gov/41000468/
2. Bordage G, Page G. The key-features approach to assess clinical decisions: validity evidence to date. Adv Health Sci Educ Theory Pract. 2018. https://pubmed.ncbi.nlm.nih.gov/29777464/
3. Bergemann AD, McMillin M, Gillispie-Taylor MC, Gingell G, Surles RT, Wallace R. Words of advice: how to write the story of a patient for a problem-based learning session in a healthcare education program. FEBS J. 2025. https://pubmed.ncbi.nlm.nih.gov/39673066/
4. Peabody JW, Luck J, Glassman P, Dresselhaus TR, Lee M. Comparison of vignettes, standardized patients, and chart abstraction: a prospective validation study of 3 methods for measuring quality. JAMA. 2000. https://pubmed.ncbi.nlm.nih.gov/10755498/
5. Peabody JW, Luck J, Glassman P, Jain S, Hansen J, Spell M, Lee M. Measuring the quality of physician practice by using clinical vignettes: a prospective validation study. Ann Intern Med. 2004. https://pubmed.ncbi.nlm.nih.gov/15545677/
6. Fournier JP, Demeester A, Charlin B. Script concordance tests: guidelines for construction. BMC Med Inform Decis Mak. 2008. https://pubmed.ncbi.nlm.nih.gov/18460199/
7. Takahashi H, Shikino K, Kondo T, Komori A, Yamada Y, Saita M, Naito T. Educational Utility of Clinical Vignettes Generated in Japanese by ChatGPT-4: Mixed Methods Study. JMIR Med Educ. 2024. https://pubmed.ncbi.nlm.nih.gov/39137031/
8. Jackson FI, Keller NA, Kouba I, Kouba W, Bracero LA, Blitz MJ. Large Language Model Clinical Vignettes and Multiple-Choice Questions for Postgraduate Medical Education. Acad Med. 2025. https://pubmed.ncbi.nlm.nih.gov/40550116/